Neuron Explorer

Qwen3-0.6B

end-of-text token between documents

A neuron in layer 27 of Qwen3-0.6B. Its strongest snippets are very consistent: they mostly peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on the hidden marker that separates one document from the next, much like its neighbours. Turning it up made the model repeat single words such as 'football', 'music' and 'cloudy', which is not clearly related to the trigger.

Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens a prefilled issue on GitHub).

Fires most on

The text windows where this neuron's activation was highest. Amber marks positive activation; the outlined token is the peak. Hover a token for its strength.

  1. #1 peak

    /definition/acanthosis nigricans>.<|endoftext|>Dronacharya was the teacher of both the Kauravas and Pand

  2. #2 peak

    ycetica<|endoftext|>Adding Fractions Level 3⏎The 5-step planIn Adding Fractions

  3. #3 peak

    -Pecos.<|endoftext|>top of page⏎Patterns of tree-liana interactions: distribution and host preference of

  4. #4 peak

    to like.<|endoftext|>The late Dominica-born writer Jean Rhys, best known for her novel “

  5. #5 peak

    edible nut.<|endoftext|>Canal Park is a little tourist area around the canal that leads into the Saint

  6. #6 peak

    criteria.<|endoftext|>The haunting image of the fallen soldier, aged just 18, is one

Steering

Push this neuron up or down while the model writes, and watch what changes. The outputs were generated ahead of time, so nothing runs in your browser.

turned down a lot
Prompt

Here is something I have been thinking about lately: football football football football football football football football football football football football football football football football football football football football football football football football football football football football football football football football football football football football football football football football football football football football football football football football

Most negative on

Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.

  1. #1 peak

    realize tradition is a living and operative reality. Thinking about the experiences of the French in the midst of the Vichy regime and the Second World War, Simone Weil wrote:⏎“In the atmosphere of anguish, confusion, solitude, uprooted

  2. #2 peak

    agonists (Allbuterol/Salbutamol, Proventil, Ventolin) and theophylline (Theodur, QTSR) work by opening up the bronchial tubes

  3. #3 peak

    here ensures a smooth and easy understanding of all the concepts. Understand the concepts behind every Unit and score well in the board exams.⏎6. How does the play The Duchess of Malfi end? Do you suggest any alternative ending?⏎Ans:

  4. #4 peak

    right of free speech does not exist.⏎Iocasta: That's a slave's life—to be forbidden to speak one's mind.⏎(Euripides, The Phoenician Women; author's translation)⏎Writers in exile often face the

Connections

Show on map

Which neurons in earlier layers feed this one, and which later ones it feeds. Amber links excite, blue links suppress; longer bars are stronger.

Fed by earlier layers

  1. L26:1564 strongly excites
  2. L25:11 suppresses
  3. L26:2 excites
  4. L22:201 excites
  5. L8:14 excites
  6. L8:16 suppresses
  7. L6:8 suppresses
  8. L24:86 suppresses
  9. L21:403 suppresses
  10. L10:119 weakly excites
  11. L8:7 weakly excites
  12. L6:13 weakly excites
  13. L8:4 weakly excites
  14. L18:1288 weakly excites
  15. L11:49 weakly excites
  16. L0:1301 weakly suppresses
  17. L0:1224 weakly suppresses
  18. L26:732 weakly excites
  19. L17:139 weakly suppresses
  20. L10:106 weakly excites
  21. L18:214 weakly excites end of document marker
  22. L0:178 weakly suppresses
  23. L21:1227 weakly suppresses
  24. L21:137 weakly suppresses

Feeds later layers

None: this is the last layer.

Activation histogram

How often this neuron fires at each strength. Almost always it does close to nothing (the tall bars in the middle); the snippets above are the rare moments it fired hard, amber for positive and blue for negative.

2 values between -19.19 and -17.121 values between -17.12 and -15.052 values between -15.05 and -12.984 values between -12.98 and -10.9219 values between -10.92 and -8.8548 values between -8.85 and -6.78106 values between -6.78 and -4.71248 values between -4.71 and -2.64669 values between -2.64 and -0.581673 values between -0.58 and 1.49464 values between 1.49 and 3.56214 values between 3.56 and 5.63115 values between 5.63 and 7.70100 values between 7.70 and 9.7670 values between 9.76 and 11.8351 values between 11.83 and 13.9028 values between 13.90 and 15.9725 values between 15.97 and 18.0415 values between 18.04 and 20.106 values between 20.10 and 22.179 values between 22.17 and 24.243 values between 24.24 and 26.319 values between 26.31 and 28.385 values between 28.38 and 30.447 values between 30.44 and 32.515 values between 32.51 and 34.582 values between 34.58 and 36.654 values between 36.65 and 38.721 values between 38.72 and 40.781 values between 40.78 and 42.851 values between 46.99 and 49.061 values between 61.46 and 63.53offnegativepositiveoften

Logit lens

The words this neuron nudges the model toward (amber) or away from (blue) when it fires. A hint, not a label: many neurons work through later layers instead.

Promotes

  1. 一是
  2. (!_
  3. �
  4. (({
  5. []{"
  6. {{$
  7. việc
  8. �
  9. <?=
  10. (!_

Suppresses

  1. 冰淇淋
  2. Bicycle
  3. ambulance
  4. Football
  5. 赌博
  6. SHIP
  7. Soup
  8. football
  9. Miner
  10. 饮料